AI to accelerate electricity connection journeys at NGED

Regression machine learning | Conformal prediction | Connections transformation

As part of an ENA Innovation challenge, National Grid Electricity Distribution wanted to improve its high voltage and extra high voltage connection journeys, using AI where it could make a meaningful difference. Baringa led a business-led AI programme that started by mapping the end-to-end process and understanding its operational challenges. We then identified the most suitable AI responses, prioritised practical opportunities and developed a proof of concept for producing budget estimates.

The challenge

The queue to connect to NGED's network had grown significantly. The pressure was greatest at HV and EHV, where generation and demand connections involve lengthy processes, complex engineering decisions and coordination across several teams.

NGED needed more than a list of AI technologies looking for applications. It needed to start with the connection journey, understand the pain points experienced by customers and planners, and determine where change would create the greatest value. Only then could it identify whether AI was the right response, which type of AI was most suitable, and what data, governance and operational changes would be required to apply it responsibly.

Our approach

We began with the business process, mapping the HV and EHV journeys with leaders, operational teams and external stakeholders. This identified 70 pain points and showed where process, policy or data changes were needed, and where AI could genuinely add value.

We assessed the AI opportunities against business value, feasibility, data readiness and delivery complexity. Nine priority AI use cases formed a phased roadmap aligned with NGED's wider connections plans and underpinned by appropriate governance.

We then built a budget estimate proof of concept. Regression models trained on historical application, network and cost data identified patterns across comparable connections enabled by exploratory data analysis, while keeping specialists in control of review and decision making.

Conformal prediction added statistically grounded cost ranges. These reflected both aleatoric uncertainty, the inherent variability in costs, and epistemic uncertainty, where the model had limited evidence. This helped users judge when an estimate was reliable and when expert review was needed, in addition to understanding data quality and additional features.

The impact

• A business-led AI roadmap: nine prioritised use cases connect specific process challenges to the most suitable AI response, giving NGED a practical sequence for investment rather than a disconnected set of technology experiments.
• More decision-useful estimates: the regression model generated an evidence-based budget estimate, while conformal prediction provided a statistically grounded range that reflected the uncertainty around it.
• Greater confidence and control: distinguishing inherent cost variability from uncertainty caused by limited model knowledge helped users judge when an estimate was dependable and when further expert planner review was needed.
• Ready for the next phase: the use cases, delivery roadmap and governance foundations give NGED a clear route towards piloting and scaling AI across connections.